Construction of ecological security evaluation model of healing landscape based on deep learning

Abstract

With the rapid growth of the global population and the increasing urbanization, the urban landscape in China is gradually enriched, and the scale of the landscape that plays a healing role is expanding. However, curing the problem of landscape ecological security is an important part of Homeland security, economic and social sustainable development. We must deal with the relationship between high-quality social development and ecological environment protection on the basis of scientific evaluation. To address this issue, research has provided better data support for feature extraction through image preprocessing. Then the Convolutional neural network in deep learning is trained through a large number of collected measured data. Finally, the pressure state response model is used to evaluate the ecological security of the healing landscape. The results show that the average error of the ground class in 2010 was 13.65%, and the fitting accuracy reached 86.35%, indicating that this method has high accuracy and can be effectively applied in evaluation. Meanwhile, in 2010 and 2019, the average landscape ecological security levels of City A were 7.27 and 6.65, both at a “safe” level, but the overall security level showed a downward trend. It is recommended to optimize the land use pattern in future urban planning and construction, improve the urban landscape ecological security index value, and maintain consistency with the actual situation of the city. This can provide reference for the evaluation model of urban landscape ecological security, and further provide scientific basis and guidance for the ecological civilization construction of urban agglomerations. In subsequent research, the evolution trend of urban landscape ecological security can be taken as the research goal, and finally, guidance on optimizing urban landscape ecological security can be provided.

Keywords

Deep learning PSR model ecological security assessment

1 Introduction

The sustainable development of human society is affected by the ecological environment, and the health and integrity of the ecological environment system can be reflected according to the degree of ecological security. The regional ecological security is the security of the regional ecological environment system, and the service function of the ecological environment system reflects the security degree of the ecological system. Ecological security refers to the ecological environment needed to maintain the sustainable development of human society. The hidden danger of ecological security caused by natural factors and human activities as well as their joint action does not pose a threat to the state or the possibility of evolution trend of human ecological security space. Urban ecological security is not only the goal of urban sustainable development, but also the important guarantee for realizing urban sustainable development. The rapid progress of urbanization has caused a lot of ecological and environmental problems, resulting in increasingly prominent hidden dangers of regional ecological security. Therefore, how to effectively and quickly assess the level of ecological security in the process of urbanization has been paid more and more attention by researchers.

The researches on ecological security evaluation at home and abroad are increasing year by year. Researchers mainly evaluate land, city, landscape, etc., and the evaluation is mainly carried out from the two aspects of index weighting and evaluation model. The index weighting is gradually objectified, and entropy weight method and combination weight method are widely used. The main evaluation models include pressure-state-response (PSR) model and its extension model, matter-element model, TOPSIS model, etc. Gong et al. [9] evaluated the ecological security of Guangzhou from the perspective of space and time, and on this basis studied the spatial differentiation and dynamic transfer characteristics of ecological security level. Based on the press-state-response (PSR) model, Li et al. [29] comprehensively applied ecological footprint and emergy theory to evaluate the current situation and development trend of ecological security in Changdao County, Shandong Province. Cao and Li [17] established the evaluation system and index of urban ecological security by emergy analysis method, and evaluated the ecological security level of Tangshan city. Based on the current research status of ecological security evaluation, Wen et al. improved and developed the principal component analysis method, combined ecological security evaluation with landscape ecology, and determined the corresponding indicator evaluation system. The results showed that it improved the accuracy of the evaluation results [8]. Ke’s team has established an urban ecological security evaluation index system based on the “driving force pressure state impact response” model. Then, the grey clustering method is introduced to construct an urban ecological security evaluation model. The results indicate that the average ecological security level of these 16 cities from 2007 to 2016 was between 0.535 and 0.647, with the Southwest region being better than the Northeast region, providing a theoretical reference for enterprise decision-making on urban sustainable development [24].

Ghosh and other scholars combined network analysis method, decision-making experiment and evaluation experiment method to assess urban ecological security, and combined cellular automata and Markov chain model to simulate land use change. The results proved the effectiveness of this method [21]. Qiu et al. constructed an ecological security evaluation index system from three aspects: pressure, governance, and environment, and studied the interactive relationship between urbanization and ecological security through grey correlation and decoupling models. The results showed that population urbanization has the greatest impact on ecological security, with a correlation value of 0.74 [14]. Wang’s team selected evaluation factors from “environmental foundation and human interference", comprehensively evaluated regional landscape ecological security through spatial principal component analysis, and constructed a regional ecological security pattern using the minimum cumulative resistance model. The results showed that the overall ecological security level of the research area was low, which has certain reference value [28]. Li et al. identified ecological patches through ecosystem services and used the minimum cumulative resistance model to derive the ecological corridor. Finally, combined with the gravity model and connectivity probability index, they analyzed the key restoration and protection areas of the ecological corridor. The results showed that this method has strong practicality [18]. Liu’s team constructed an evaluation index system based on the supply and demand of ecosystem services in the pressure state response model, evaluated the ecological security of the Pearl River Delta, and identified the main obstacles to ecological security. The results showed that in addition to grain yield and habitat quality, soil retention capacity, carbon sequestration, and water yield increased with fluctuations, achieving the screening of key indicators [23]. Overall, current researchers are focusing on how to select and extract appropriate features for the ecological safety evaluation task of recuperation landscapes, in order to obtain better evaluation results. At the same time, researchers have established a PSR model to integrate and analyze stress factors, ecological environment status, and response measures, evaluate the health status and ecological security risks of the ecosystem, and achieved good results. However, there is still a lack of corresponding method support in accurately identifying pressure factors, selecting and measuring ecological environment status indicators, and evaluating the effectiveness of response measures. Moreover, research on the ecological safety evaluation model of recuperation landscapes based on deep learning is still in its infancy.

As urban ecosystem tends to be complex, heterogeneous and discontinuous, and various technologies become mature, principal component analysis, fuzzy system, GIS, obstacle degree model and PSR model are used more and more, which provides more accurate quantitative technical means for regional ecological security research. At the same time, the prediction of ecological security can effectively guide the construction of regional ecological environment, provide reference for relevant management and decision-making departments, and make the ecological security state of the study area develop in a good direction. In general, the existing ecological security system evaluation involves different scales, and the meso-scale evaluation research (such as river basin, region) is the core of ecological security research [20]. Current research results can reflect the status of urban ecological security to some extent, but the selected index system is mostly based on statistical yearbook data, ignoring the impact of urban heat island caused by rapid urban development on urban ecological security, and failing to deeply explore various information and indicators contained in remote sensing images. The research shows that the factors and scales involved in urban ecological security are complicated, so it is necessary to construct a more comprehensive evaluation index system that can better reflect the status of urban security by combining the characteristics of typical cities.

As one of the cities with rapid economic development, the type of urban land use in A city has undergone great changes in recent years, and the internal structure and function of the regional landscape ecosystem are facing great pressure. Taking A city as an example, this paper established an effective landscape ecological security evaluation system based on PSR conceptual model, combined with remote sensing data and GIS technology, focused on the ecological security status of A City during 2010-2019, and revealed the changes and rules of landscape ecological security status of A City through inter-regional and inter-annual comparison. Through intelligent evaluation of the ecological safety of healing landscapes, the research aims to provide decision-making support for ecological protection and further promote sustainable urban development, achieving the establishment of an environmentally friendly society in the construction of digital cities.

2 Research methods

The research technology scheme of this paper includes three parts: high-resolution satellite image preprocessing, data classification based on deep learning and ecological security assessment based on PSR model.

2.1 Satellite data preprocessing

In the process of acquiring remote sensing data, it is impossible to record the surface information accurately. Due to the influence of the system itself, space and other aspects, some errors inevitably exist, and these errors reduce the quality of remote sensing image. Therefore, before using the image for analysis, the original remote sensing image must be preprocessed. The main purpose of image preprocessing is to correct geometric and radiative deformation in the original image. The process of optical satellite data processing generally includes radiometric calibration, calculation of apparent reflectance, cloud removal, atmospheric correction, image registration, fusion, Mosaic, cropping and so on. The processed data can be directly used for feature extraction and feature selection, and land use classification of remote sensing images.

2.2 Classification of deep learning

Machine learning, as the basic principle of convolutional neural networks, aims and focuses on making computers simulate how humans acquire new knowledge and abilities. The concept of deep learning is to build a continuous hierarchy of computer models that constantly learn abstract and high-level features from simple combinations of concepts. During this period, the perceptron in the hidden layer is invoked instead of manual information retrieval and information classification. Deep learning starts from the motivation of building a neural network simulating human brain to analyze and learn, and focuses on identifying and analyzing the characteristics of the target [5].

The basis of deep learning and brain analysis is the computational element, whose unit nerves are activated and become active when it receives input data. Many of these neurons are then combined and linked together to form a neural network. Neural network is a kind of mathematical model which can realize distributed and parallel information processing. It is characterized by simulating human brain neurons through artificial neural network. The most powerful computing unit in neural networks has the ability to weight input and output information. The main process is the formation of arithmetic units by using activation functions

The input-output information is transmitted to the next neuron and becomes the input-output of the next neuron.

Deep learning is a new research direction of machine learning. It mainly uses multi-layer nonlinear structure to extract data features layer by layer from low to high. The deep learning model consists of multi-layer perceptron such as input layer, hidden layer and output layer. Like machine learning methods, deep machine learning methods can be divided into supervised learning and unsupervised learning. Different learning frameworks have different learning models. The learning framework adopted in this paper is the TensorFlow framework, and the established learning model is Convolutional Neural Network (CNN), which is a machine learning model under deep supervised learning [12]. The typical CNN structure is composed of input layer, convolution layer, pooling layer and fully connected layer. Each layer has multiple feature maps, and each feature map extracts a feature of the input through a convolution filter.

(1) Input layer

The input layer is composed of neurons, which act as a transmission bridge and can transmit input data to other layers. The number of elements in the input layer is equal to the number of variables in the data set. The hidden layer between the input layer and the output layer is a Convolutional Neural network (CNN for short) that computes the input data. The biggest advantage of CNN is that it can recognize specific visual rules starting from the original image with little preprocessing, so it is widely used. CNN can identify all kinds of distorted and invariant 2D images very friendly. The most obvious difference between CNN and other deep learning models is weight sharing and local connection, which is also one of the most important characteristics of CNN.

(2) Convolutional layer:

Multiple convolution kernels constitute the convolutional layer, which can extract different features of the image through convolution operation, obtain lower-level image attribute features from the previous convolution layer, iterate layer by layer, and obtain complete and complex attribute information of the image from the underlying features. In order to extract multiple feature maps with multiple neural nodes, the convolutional layer mobilizes the convolutional kernel to actively connect neurons, which connects the neurons in the lower layer with a local neuron in the feature map of the upper layer, so as to achieve the purpose of extracting local features of the input image.

Each convolution kernel corresponds to a function, which is responsible for defining the weight of the weighted average value during the change process from pixels in a certain area of the input image to each pixel in the output image. The number, size and step size of the convolution kernel correspondingly determine the number, size and density of the generated feature maps. In order to avoid the phenomenon of fitting when the operation is too complex, the complexity of image data set should be fully considered in the process of determining the number of appropriate convolution kernels. Weight sharing can greatly reduce model complexity.

(3) Activation function

The two main states of neurons in neural network are activated state and inhibited state respectively, and corresponding neurons are active state and inactive state. Activation function enables neurons to acquire self-learning and adaptive ability.

(4) Pooling layer

The pooling layer, also known as the feature mapping layer, is the process of downsampling. The feature training process of CNN convolutional layer extraction is complex and computatively intensive. It can reduce the feature dimension, reduce the complexity, improve the training speed and calculation efficiency, and help prevent overfitting.

(5) Fully connected layer

After multiple convolution and pooling, the dimension of the input feature is reduced to the point where it can be directly processed by the feedforward network.

In the fully connected layer, all the features of the two-dimensional image are spliced into one-dimensional features and input into the fully connected layer.

The training of CNN is carried out through back propagation and stochastic gradient descent, and the weight and bias of convolutional neural network are updated layer by layer. A large number of measured data are collected in this paper, which establishes a solid data foundation for classification using CNN. This paper uses sample data for CNN model training. Since CNN can well obtain texture features and pixel space features of remote sensing images, model training and optimization of model parameters can further improve the accuracy of model classification. Land classification is carried out on remote sensing images, and feature extraction is carried out on the ground objects in the images by using the convolutional layer in the convolutional neural network model, so as to effectively extract the deep features, find the law of features possessed by the ground objects, and map the original image into the feature space of the hidden layer [3]. Then the distributed feature mapping of the extracted ground object image is expressed into the marking space of the ground object image sample through the neural network.

For deep learning classification results, there are two Tensor Board model accuracy curves; The other is the confusion matrix.

Tensor Board is a visualisation tool for TensorFlow that allows you to understand the overall performance and state of a classification model from the log files it has output as it learns [7, 22]. The training of a neural network is very complex. A Tensor Board is an effective way to show the calculation graphs of the training model, the change trends of various indicators over time and the images used in the training. Through comparing the hit ratio of the training set and the hit ratio curve of the verification set, the overall performance of the model will be evaluated.

Confusion matrix is especially suitable for supervised learning. The function of confusion matrix is to observe the performance of the model in various categories, and the accuracy of the model corresponding to each category can be calculated. Through confusion matrix, it can also directly observe which categories are not easy to distinguish, for example, how many categories in category A are classified into category B, so that the design features can be targeted. It makes the categories more distinct.

2.3 Ecological security assessment

2.3.1 Selection of evaluation model

Landscape ecological security evaluation is a comprehensive system engineering, and the evaluation system needs the cross combination of several disciplines.

This paper uses the pressure - state - response (PRS) model as the landscape ecological security assessment model of Hengshui Lake wetland. In the PRS model, can directly reflect the interaction and connection between human and environment, that is, human activities on the environment pressure, the indicators and system in the environment will change the original structure and state, according to the change of the environment, human will also through their own economic and social activities or management strategies to respond to the change of the environment, In order to prevent environmental degradation to human harm and crisis.

2.3.2 Principles of evaluation index system construction

(1) Scientific principle. The selected indicators should have clear ecological significance, follow the law of sustainable development, and objectively reflect the complexity and diversity of the ecological environment in the study area.

(2) The principle of dynamism. Ecological security is dynamic and long-term, so it is necessary to consider the change value in different periods and the influence of spatial factors in the selection of indicators.

(3) Operability principle. The selected indicators can not only combine orientation and quantification, but also reflect the state of ecological security more accurately and comprehensively.

2.3.3 Construction of evaluation index system

In this study, it is expected to use analytic hierarchy process to construct the evaluation index system. The ecological security evaluation index system is constructed from the objective layer, criterion layer and index layer, and the objective of this evaluation system is landscape ecological safe water

Where the criterion layer is composed of pressure (B1), state (B2) and response (B3) systems in the PSR model.

3 Case study

3.1 Overview of the study area

A City is an important city in central China, with a total land area of 4996 km². Coniferous broad-leaved mixed forest area, plant community is diverse. The terrain is high in the southwest, low in the northeast, and varied in topography. From south to north, it is mountainous, hilly, ridge and plain respectively. The southern low mountain area is dominated by grass and shrub vegetation. The central and southern hilly area is dominated by masson’s pine, Chinese fir forest and pine and oak mixed forest. The forest species in north-central Longgang area is the same as that in south-central hilly area. The plain area along the river belongs to wet - growing and semi-wet - growing vegetation area. Mostly for roads, canals, rivers, village side artificial scattered timber forest, and a small amount of desert fruit forest and agroforestry intercropping, forest coverage rate of 14.6%. As one of the cities with rapid economic development, the type and structure of land use in A City also change with the development of time, and the internal structure and function of the regional landscape ecosystem are also facing great pressure.

3.2 Data source and preprocessing

This project uses Planet data from the United States to carry out ecological monitoring and analysis of the whole research area. Planet satellite is the only commercial remote sensing satellite in the world with global high-resolution, high frequency and full coverage capability. It has the following characteristics: high data coverage efficiency. Autonomous coverage of images. Planet satellite images do not need to be programmed. Hundreds of satellites take autonomous pictures of the world every day.

Landsat TM/ETM + remote sensing data from October 21, 2010 and December 19, 2019 were selected for the study. A total of 18 satellite images (maps) of 5 square kilometers (5×5 km in size) were screened. The image size is about 9000×9000 pixels, and the file size is about 400 MB. The printed image is about 3×3 meters in size when printed at the ordinary resolution of 72 DPI. First of all, ENVI software was used for radiometric correction of the images, and then image-to-image registration was adopted to register the images of 2019 with the images of 2010 as the basis. 25 control points are uniformly selected on the image, and the error after registration is 0.49 pixels.

3.3 PSR model construction

3.3.1 Construction of landscape ecological security evaluation index

Based on the principles of scientificity and practicality, operability and comparability, comprehensiveness and dominance, combination of quantitative and qualitative indicators, sustainable development and dynamics. At the same time, according to the relevant standards stipulated by local, industrial and national regulations, background and background standards, analogy standards, ecological effects determined by scientific research, target standards in ecological construction, expert experience values and other relevant standards, the ecological security evaluation system is finally determined to be composed of three levels of target layer, criterion layer and index layer and 7 selected indicators.

The landscape ecological security pattern is composed of some key parts, locations and spatial connections of landscape which are important for controlling ecological processes. As an entry point in the study of ecological security, in the formation and evolution mechanism of landscape ecological security, the PSR conceptual model proposed by the United Nations Economic Development and Cooperation Agency (OECD) was applied to construct the landscape ecological security assessment system in A city [2, 4, 19, 26, 27]. The evaluation system consists of target layer, criterion layer and index layer, among which criterion layer includes pressure layer, state layer and response layer. Each criterion layer contains several indicators, and the selected indicators should meet the requirements of comparability, accessibility and quantification as far as possible (see Table 1).

Table 1
Index system of landscape ecological security assessment

Target layer Criterion layer Index layer

A Comprehensive B1 Pressure layer C1 Population pressure

index of landscape C2 Intensity of urban expansion

ecological security B2 State layer C3 Vegetation index

C4 Ecological elasticity

C5 Biological richness

B3 Response layer C6 Landscape structure index

C7 Intensity of heat island effect

Target layer	Criterion layer	Index layer
A Comprehensive	B1 Pressure layer	C1 Population pressure
index of landscape		C2 Intensity of urban expansion
ecological security	B2 State layer	C3 Vegetation index
		C4 Ecological elasticity
		C5 Biological richness
	B3 Response layer	C6 Landscape structure index
		C7 Intensity of heat island effect

(1) Landscape pressure layer (B1) index

Human activities and natural disasters bring potential pressure to the stability of natural ecosystem and promote the change of landscape pattern. The expansion of urban land and the rapid growth of population are the main pressures of landscape ecological security in the process of urbanization in A City. Therefore, the disturbance intensity of landscape ecosystem security can be expressed by population pressure and building land index. The distribution of population pressure can be understood as taking urban area as the point source pressure center and decreasing step by step with the increase of outward distance. The range of urban main urban area can be obtained by visual interpretation of remote sensing image and buffer analysis [6, 10, 15]. The reflectance of urban building land in the middle infrared band is higher than that in the near infrared band, so the building land index (NDBI) can be used to represent the urban land expansion status, and its calculation formula is as follows: $NDBI = \frac{ρ_{MIR} - ρ_{NIR}}{ρ_{MIR} - ρ_{NIR}}$ (1)

Where, ρ_MIR and ρ_NIR are the reflectance or brightness values of image elements in mid-infrared band and near-infrared band, respectively, representing the pixel brightness values of TM/ETM+ 5,4 bands respectively.

(2) Landscape state layer (B2) index

The normalized vegetation index (NDVI), ecological elasticity (ECO) and biological abundance index (BAI) were selected to reflect the state of landscape ecosystem from three perspectives: landscape vitality, component structure and ecological contribution [5].

Normalized vegetation index (NDVI). NDVI was used to quantitatively evaluate the vegetation coverage and growth vigor in the study area, and its calculation formula was as follows: $NDVI = \frac{ρ_{NIR} - ρ_{Red}}{ρ_{NIR} + ρ_{Red}}$ (2)

Where, ρ_NIR and ρ_Red are the reflectance or brightness values of image elements in near-infrared band and red band, respectively, representing the pixel brightness values of TM/ETM+ 4,3 bands respectively.

Ecological resilience (ECO). ECO refers to the degree to which the ecosystem can recover to the original state after deviating from the original state when the internal and external disturbance or pressure of the ecosystem does not exceed its elastic limit. It is used to represent the buffering and adjustment ability of the ecosystem. The method of extracting ECO from ecosystem health assessment is applied to landscape ecological security assessment, and the calculation formula is as follows: ${ECO}_{res} = \sum_{i = 1}^{n} (S_{i} • P_{i})$ (3)

Where, S_i is the area of class i land use type; P_iis the elastic score of Class i land use type; n is the number of land use types.

Biological abundance index (BAI). BAI reflects the richness and poverty of biodiversity (biological richness) in the evaluated region, and the formula is as follows: $BAI = A_{bio} \frac{\sum_{i = 1}^{n} (S_{i} • P_{i})}{S}$ (4)

Where, A_bio is the normalization coefficient; S_i is the area of Class i land use type; P_i is the biological abundance weight of Class i land use type; S is the total area of the region; n is the number of land use types.

(3) Landscape response layer (B3) index

Landscape response refers to the reflection of landscape to the current environmental pressure. Or the positive compensatory measures taken by human perception after the regional ecological security problem is affected. Landscape structure index can quantitatively describe the composition and configuration characteristics of landscape spatial pattern structure in the study area, which is represented by the linear combination of landscape fragmentation, diversity, fractal dimension and vulnerability. Therefore, the study of regional landscape ecological structure can reveal the regional ecological security status. Urban thermal environment is an important index of urban ecological environment, and its spatial distribution characteristics are the result of interaction between human activities and urban landscape. By integrating the thermal environment into the theory of landscape ecology and studying the spatio-temporal process of urban thermal landscape, the ecological security of urban landscape can be indirectly reflected.

At present, most of the research on urban thermal environment focuses on the spatial distribution of temperature relative intensity. Therefore, this study uses ground radiant brightness temperature to study the thermal environment of A city.

The formula of ground radiation brightness temperature inversion using Landsat TM/ETM + data is as follows [11]. $T = K_{2} / ln (K_{1} / L + 1)$ (5)

Where, T is radiant brightness temperature, K; L is the radiation intensity of thermal infrared band; K₁ and K₂ are the correction coefficients.

3.3.2 Determination of the weight of landscape ecological security evaluation index

According to the unique geographical characteristics of the study area and the availability of data, Delphi method is used in this study to determine the weight value of each ecological security evaluation index. The research has carried out several opinions of experts, who come from forestry, environmental protection, water conservancy, national land, university research institutions and other related industries. A total of 12 experts, scholars, managers and professors who know more about the research area or the industry are selected. According to the opinions and suggestions of experts in different industries, the final weight value of the index is processed by mathematical formula. In addition, each weight value is given a 10-point scoring system to simplify the calculation.

Landscape ecological security evaluation factors reflect the relative importance of each index in the evaluation system and affect the evaluation result of the whole ecological security. According to the actual situation of the study area, AHP is used to determine the weight of evaluation factors. The index weights obtained by this method are objective and accurate. After calculation, the weight of each layer is shown in Table 2. The random consistency ratio of the target layer of the judgment matrix is 0.0043 (less than 0.1).

Table 2
Weight of ecological security assessment index system

Target layer Criterion layer Index layer Weight w_i

A B1(0.2279) C1(0.0915) 0.4013

C2(0.1364) 0.5987

B2(0.4745) C3(0.2711) 0.5713

C4(0.0816) 0.2567

C5(0.1218) 0.1721

B3(0.2976) C6(0.1194) 0.4013

C7(0.1782) 0.5987

Target layer	Criterion layer	Index layer	Weight w_i
A	B1(0.2279)	C1(0.0915)	0.4013
		C2(0.1364)	0.5987
	B2(0.4745)	C3(0.2711)	0.5713
		C4(0.0816)	0.2567
		C5(0.1218)	0.1721
	B3(0.2976)	C6(0.1194)	0.4013
		C7(0.1782)	0.5987

3.3.3 Standardization of landscape ecological security evaluation indexes

The evaluation value dimensions of the index factors obtained by the above methods are not uniform and are not comparable. Therefore, the original data index factors are normalized to make the final standard quantization value between (0 ∼ 10).

The index processing method that the larger the value is, the safer the ecology is: $P_{i} = \frac{X_{i} - X_{min}}{X_{max} - X_{min}} \times 10$ (6)

Where, X_i (i = 1, 2, ⋯ , n) is the original value of the ith evaluation index; X_max is the measured maximum value; X_min is the minimum measured value; P_i is the standardized value [26].

The index processing method that the smaller the value is, the safer the ecology is: $P_{i} = (1 - \frac{X_{i} - X_{min}}{X_{max} - X_{min}}) \times 10$ (7)

Where, X_i (i = 1, 2, ⋯ , n) is the original value of the ith evaluation index; X_max is the measured maximum value; X_min is the minimum measured value; P_i is the standardized value.

3.3.4 Calculation of comprehensive evaluation index of landscape ecological security

The comprehensive evaluation index of landscape ecological security is obtained by weighted summation of each index factor, and the evaluation model is as follows [10]. $LESI = \sum_{i = 1}^{n} (w_{i} • x_{i})$ (8)

Where, LESI is the evaluation index of landscape ecological security, and x_i is the evaluation vector of the ith evaluation factor; w_i is the weight vector of the ith evaluation factor; n is the number of evaluation indicators.

According to the actual situation of the study area and combined with the classification methods of landscape ecological security levels at home and abroad, landscape ecological security index values and security levels were associated by non-equidistant division, as shown in Table 3.

Table 3

Criterion of integrative index of landscape ecological security assessment

LESI	<3.8	[3.8, 4.8)	[4.8, 6.5)	[6.5, 8.5)	⩾8.5
Level	Pathological	Not safe	Critical	Safe	Very safe

In terms of calculating the comprehensive index of urban landscape ecological security evaluation rthe PSR model index system weight is first obtained using the entropy weight principal component analysis method. Then rthe results obtained using the comprehensive index method are used to obtain the urban landscape ecological security evaluation comprehensive index. The calculation formula is shown in Equation (9). $U = \sum_{i = 1}^{m} w_{i} \cdot S_{ij}$ (9)

In Equation (9) rS_ij represents the safety index score of each principal component indicator rand w_i represents the weight of the indicator items in each indicator layer. The indicator safety index adopts the standard score method rand is divided into positive and negative indicators based on the trend of the indicators for ecological safety index calculation. For positive indicators rthe urban landscape ecological security index is calculated as shown in Equation (10). ${\begin{matrix} x_{ij} ⩾ r_{i}^{'} r S_{ij} = 1 \\ x_{ij} < r_{i}^{'} r S_{ij} = \frac{x_{ij}}{r_{i}^{'}} \end{matrix}$ (10)

For reverse indicators rthe urban landscape ecological security index is calculated as shown in Equation (11). ${\begin{matrix} x_{ij} > r_{i}^{'} r S_{ij} = \frac{r_{i}^{'}}{x_{ij}} \\ x_{ij} ⩽ r_{i}^{'} r S_{ij} = 1 \end{matrix}$ (11)

In Equations (10) and (11) rx_ij is the original value of the i principal component of study year j; r_i is the benchmark value of the evaluation indicator (set as the average value of each indicator within the study year); $r_{i}^{'}$ is the value obtained from the principal component method dimensionality reduction of the benchmark value of the evaluation index. This step requires data standardization and the principal component method dimensionality reduction. The benchmark value r_i should be considered and calculated synchronously with the original indicator data.

4 Experimental results and analysis

4.1 Precision of deep learning model

The deep learning module of ENVI software is studied for classification. Accuracy and user accuracy are expressed as the accuracy of the deep learning training model, ranging from 0 to 1. The closer the value is to 1, the better the classification effect will be. In the research, the accuracy of the deep learning model and the user’s accuracy increased with the increase of the number of iterations. The input layer samples are standardized data, which is dimensionality reduced through principal component analysis, which can reduce training costs and improve training accuracy. The number of nodes in the input layer is 4. The output layer sample is the comprehensive evaluation index obtained by entropy weight principal component analysis method, and the number of nodes in the output layer is set to 1. At the same time, the super parameters of the network are set. The Learning rate of the network is set to 0.004, and the momentum factor is set to 0.99.The parameter settings of the deep learning model used in the study are shown in Table 4.

Table 4
Structure and parameter setting table of the model

Network layer Settings

Convolution Layer Conv2d(1,3,(3,51))

Attention Layer Conv2d(2, 3, (3, 51), padding = 0, bias = Flase); Sigmoid()

Pooling Layer MaxPool2d((2, 1), 1)

Fully Connected Layer Linear(51, 4)

Classify Function Softmax

Network layer	Settings
Convolution Layer	Conv2d(1,3,(3,51))
Attention Layer	Conv2d(2, 3, (3, 51), padding = 0, bias = Flase); Sigmoid()
Pooling Layer	MaxPool2d((2, 1), 1)
Fully Connected Layer	Linear(51, 4)
Classify Function	Softmax

After 27 iterations, the model’s accuracy was high, meeting the research needs. The range of loss value (loss) is also 0-1, but the trend of loss value is opposite to the trend of model accuracy. Loss value tends to 0, indicating the higher degree of matching, and vice versa, the worse degree of matching. After 27 iterations, the model’s matching degree meets the research needs.

Through statistical analysis of class accuracy and classification confusion matrix, the overall classification accuracy in 2010 was 85.74%, Kappa coefficient was 0.8483, and all kinds of ground objects were well recognized. In 2019, the overall classification accuracy was 84.88% and Kappa coefficient was 0.8410, which slightly decreased compared with 2010.

In order to evaluate the actual accuracy of the deep learning classification, 2000 random points were selected by ArcGIS and identified and classified through high-definition Google images. Field tests were carried out on the images that were not easy to be identified, and then the error coefficient was calculated [1, 16, 25].

In 2010, the average error of ground class was 13.65%, and the fitting accuracy reached 86.35%. The average error of the ground class in 2019 was 16.40%, and the fitting accuracy reached 83.60%. At the same time, Root-mean-square deviation (RMSE) and Mean absolute error (MAE) are indicators to measure the accuracy of the algorithm [13]. The results show that the deep learning classification method has a mean MAE of 0.793 and a mean RMSE of 0.628, indicating a lower error rate. It can be seen that deep learning classification is a feasible and effective method to identify ground object types.

4.2 Ecological security assessment of landscape types

Firstly, compare the proposed method with the methods in the literature review, including principal component analysis, minimum cumulative resistance model, etc., to evaluate the selected urban area. The remote sensing images used were Landsat MSS and TM images from 2010, 2013, 2015, 2018, and 2019. The images obtained should be selected from June to September with good vegetation cover and no or less clouds, in order to better judge and interpret the image. At the same time, the study referred to the official statistical data of the city, and the remote sensing image preprocessing was carried out using ENVI software. These data are relatively simple and reliable to obtain, and are all obtained through GIS, GPS, and RS technologies. To eliminate interference from other indicators, the selected evaluation range and indicators are consistent. The comparison results between the proposed method and the methods in references [7, 8, 12, 14, 18, 21–24, 28], are shown in Table 5. From Table 5, it can be observed that the evaluation accuracy of the proposed method is 85.7%, with an average time consumption of only 37 seconds, which is superior to other methods and has high accuracy and efficiency.

Table 5
Evaluation accuracy and time-consuming results of 11 methods

Method type Accuracy/% Time consuming/s

Reference [8] 68.5 45

Reference [24] 73.7 58

Reference [21] 67.3 42

Reference [14] 58.1 63

Reference [28] 70.9 47

Reference [18] 71.8 69

Reference [23] 75.7 68

Reference [12] 67.1 44

Reference [22] 63.2 40

Reference [7] 70.7 45

Proposed method 85.7 37

Method type	Accuracy/%	Time consuming/s
Reference [8]	68.5	45
Reference [24]	73.7	58
Reference [21]	67.3	42
Reference [14]	58.1	63
Reference [28]	70.9	47
Reference [18]	71.8	69
Reference [23]	75.7	68
Reference [12]	67.1	44
Reference [22]	63.2	40
Reference [7]	70.7	45
Proposed method	85.7	37

The comprehensive index map of landscape ecological security in Longyan City was obtained by using FRASTATS software and ArcGIS spatial analysis module (Fig. 1).

Fig. 1

Spatial distribution of landscape ecological security index.

The land use classification maps of the two periods were superimposed with the landscape ecological security assessment images of the same period respectively, and the mean values of ecological security of each landscape type were calculated in 2010 and 2019.

It can be seen from Fig. 2 that forest land and water were at the ecological security level in 2010. The value of construction land and agricultural land is between 4.5 and 5, which is in the critical state of safety. There are some hidden dangers of ecological security in unused land. In 2019, the average ecological security of forest land and water body was 6.97 and 5.52 respectively, while agricultural land was the lowest among all landscape types.

Fig. 2

Mean-line graph of landscape ecological security.

In the past 10 years, except the ecological security level of water body increased, the security value of other landscape types decreased, especially the woodland landscape. The main reason is that in the process of urbanization in A city, the area of urban building land continues to expand, encroaching on the urban green space and the surrounding cultivated land. In addition, the road network cuts the woodland landscape, which reduces the security level of woodland and agricultural land. Due to the increase of building land, the urban vegetation is relatively reduced, and the ecological security of building land is also decreased. On the other hand, the water landscape in A city is at the critical safety level, and the overall level has been improved slightly. It can be seen that the effect of water treatment in Longyan City during this period has initially appeared, but it is not optimistic. The consequences of this will be an increase in urban ecological environmental pressure, worsening pollution, damage to urban economic development, and a serious threat to sustainable development capabilities.

4.3 Comprehensive index analysis of landscape ecological security

The statistical analysis of the spatial distribution of landscape ecological security index in Longyan City shows that the average value of landscape ecological security in A City in 2010 is 7.27, which is at a high security level. The average in 2019 was 6.65, which is the medium safe level. The result reflects that the landscape ecological security in A city is still in a safe state, but the overall level has a small decrease. The respective ecological security levels of the two periods were statistically analyzed, and Fig. 3 was obtained.

Fig. 3

Raking percentage of landscape ecological security assessment.

As can be seen from Fig. 3, Among the two periods’ respective safety levels, the proportion of “safety” level is the highest, exceeding 50% of the total area of the research area. Based on the analysis in Fig. 1, the safety zone corresponds to a large area of forest land on the outskirts of the city. It shows that the higher forest coverage in Longyan has the greatest contribution to the ecological security of the whole region. In the ecological security level, the proportion of “pathological” and “very safe” levels is relatively small. Based on the analysis of Fig. 1, it is believed that the “very safe” level has the largest change, with a decrease of 7.74%, and the decrease mainly corresponds to the woodland near the urban area. Between 2010 and 2019, the weights of “sick", “unsafe", and “critical” in the ecological security level all increased, reaching 0.56%, 4.37%, and 5.31%, respectively. In 2010, most of the areas with a “safe” ecological security level transformed into a “critical” state, corresponding to the urban construction land area. In order to analyze the change of landscape ecological security in Longyan from a macro perspective, the landscape ecological security change curve was drawn, as shown in Fig. 4.

Fig. 4

Change curve of landscape ecological security in Longyan.

The two major peaks of the two curves in Fig. 4 are roughly located between the ecological security values of 7.5 and 8.5, and the peaks in 2010 have significantly decreased, indicating a trend of ecological security moving towards low values. At the same time, there are also two small peaks in the range of 4∼6 of the ecological security value, of which the peak in 2010 is significantly higher than that in 2019, which indicates that the ecological security level of most areas in Longyan was higher in 2010 and 2019, but by 2019, the landscape ecological security degree had declined. At the same time, a considerable portion of the regional ecological security comprehensive value hovers around 5, approaching the “critical safety value” level. It can be seen from Fig. 1 that the areas with higher ecological security level in Longyan are mainly located in Jiangshan Township, which is far away from the built-up area and less disturbed by human beings. The main urban area of Xinluo District, Longyan City, with Tianma Mountain and Denggao Park as the center, is blue, which belongs to the critical safety state, and a small part of it is purple, indicating that the ecological safety level has changed to ecological deterioration, reminding relevant departments to pay attention to it.

5 Conclusion

Based on the “press-state-response (PSR)” ecological security assessment model, this paper evaluates and analyzes the ecological security situation of A city. The results objectively reflect the ecological security situation of the study area, and provide ecological security warning information for relevant departments to maintain the ecological environment of the city. Reduce the negative impact of urban expansion on the ecosystem.

The research results showed that the average landscape ecological security degree of A city in 2010 and 2019 was 7.27 and 6.65 respectively, both of which were in the “safety” level. However, the increase of landscape pressure index and the decline of ecological state index led to A declining trend of landscape ecological security level of A city. The proposed method can comprehensively evaluate the ecological security of cured landscapes, providing a more scientific and rigorous model for the evaluation of urban landscape ecological security, and providing technical support for the macro regulation of urban landscape ecological security. Therefore, in the future urban development planning and construction of A city, emphasis should be placed on protecting the forest land around the city to gradually reduce the pressure on the ecological environment. It is specifically reflected in the optimization of the urban center area with low ecological security degree and the protection of the area with high ecological security degree, so as to realize the healthy and safe development of the ecological environment of the whole city A. At the same time, in terms of land use, strict implementation of the overall land use plan. Strengthen the evaluation of intensive land use, and link the allocation of planning and control indicators with intensive land use. Gradually improve the standards for construction land indicators, strengthen the management of construction land indicators, and increase the area of landscape construction land. In addition, the land resource utilization and environmental issues brought about by the urbanization process are important obstacles to the sustainable development of cities. Therefore, promoting the combination of urbanization and green development is an important strategy for optimizing the utilization of urban land resources. In urban planning, the layout of urban space should be reasonable to avoid overcrowding and disorderly expansion. At the same time, attention should be paid to the construction of green buildings, ecological transportation, low-carbon cities, and other aspects to improve the quality of urban ecological environment. Public participation is crucial in the utilization and planning management of urban land resources. Public participation can strengthen the planning and supervision of land resources in cities, and prevent negative development behaviors. In addition, public participation can also promote the sustainable use of urban land resources and environmental protection, and improve the overall development level of the city.

The rapid development of cities and the continuous expansion of urban space have led to a rapid increase in construction land landscapes, a rapid decrease in landscapes such as arable land, forest land, grasslands, and water bodies, and the development of unused land. The number of patches in various landscape types is constantly increasing, especially in construction land and arable land, which exacerbates the fragmentation and heterogeneity of the landscape. The intensity of human interference in the entire landscape continues to deepen, and the landscape tends to become diversified and complex, resulting in drastic changes in landscape patterns. The ecological system in the research area is becoming more fragile, and ecological security is constantly decreasing. Ecological insecurity risks are gradually increasing and constantly emerging. Therefore, the inspiration for other cities facing similar problems is that in the future, they should continuously strengthen the protection and management of the landscape, control the disorderly expansion of the city, reduce the interference and damage of human activities on the natural landscape, and reduce the pressure of urban expansion on the surrounding landscape ecology. At the same time, improve the stability of the landscape structure, gradually increase the level of ecological safety in the landscape, gradually eliminate the hidden dangers that lead to ecological insecurity in the landscape, and ensure that the level of ecological safety in the landscape is always protected in a good state.

The factors and scales involved in urban ecological security evaluation are complicated, and the selection of indicators and weight assignment are the core contents of ecological security evaluation. The evaluation model constructed in this paper can objectively reflect the ecological security situation of the study area, but the determination of index weight is subject to the influence of expert experience and knowledge, which has a certain subjectivity. Therefore, the selection and weighting of indicators still need further research in the future work. At the same time, “3S” (GIS, GPS, RS) technology is also an important means of studying urban landscape ecological security. Later research can consider combining the data obtained from “3S” technology with existing data to optimize the evaluation index model when constructing the evaluation index model. In addition, deep learning can not only study the current situation of urban landscape ecological security, but also predict the evolution trend of urban landscape ecological security through algorithms. In subsequent research, the evolution trend of urban landscape ecological security can be taken as the research goal, and finally, guidance on optimizing urban landscape ecological security can be provided. In terms of optimizing technology and landscape ecological security construction, build targeted urban landscape ecological security protection strategies, narrow the gap, improve the overall level of urban landscape ecological security, and promote high-quality regional development. In the field of ecological protection and sustainable development, intelligent evaluation of cured ecological landscapes can more accurately collect information related to ecological security, provide decision-making support for ecological protection, and lay a technical foundation for the sustainable development of cities.

Footnotes

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Conflicts of interest

The authors declare that there are no conflict of interest regarding the publication of this paper.

References

Cheng

, Zhu

, Zhao

and Chen

, Situation-aware IoT service coordination using the event-driven SOA paradigm, IEEE Transactions on Network and Service Management 13(2) (2016), 349–361.

, Quan

, Yang

and Dong

, A social-ecological coupling model for evaluating the human-water relationship in basins within the Budyko framework, Journal of Hydrology 619 (2023), 129361.

Yuan

C.X.

, Xie

Y.W.

, Zhao

C.Y.

and Feng

Z.D.

, Establishment and application of NWA system in watershed ecosystem quality assessment based on RS and GIS, 2009 IEEE International Geoscience and Remote Sensing Symposium, 12-17 July 2009, Cape Town, South Africa. IEEE 2010, pp. II638–II-641.

, Zhu

, Deng

, Sun

, Ma

and Liu

, Evaluation and empirical study of Happy River on the basis of AHP: A case study of Shaoxing City (Zhejiang, China), Marine and Freshwater Research 74(10) (2023).

Wang

and Gu

, Impact of ecological security on urban sustainability in Western China—A case study of Xi’an, Environment and Planning B: Urban Analytics and City Science 48(5) (2021), 1314–1339.

Chen

G.B.

and Chen

Z.S.

, Regional classification of urban land use based on fuzzy rough set in remote sensing images, Journal of Intelligent & Fuzzy Systems 38(4) (2020), 3803–3812.

Shang

and Zhang

, Regional ecological security assessment of Dalian based on PSR model, 2017 Annual Conference of Science and Technology of Chinese Society for Environmental Sciences, Xiamen, China. Chinese Society of Environmental Sciences 2017, pp. 4096–4105.

Wen

and Hou

, Research on the progress of regional ecological security evaluation and optimization of its common limitations, Ecological Indicators 127 (2021), 107797.

Gong

J.Z.

, Liu

Y.S.

, Xia

B.C.

and Zhao

G.W.

, Urban ecological security assessment and forecasting, based on a cellular automata model: A case study of Guangzhou, China, Ecological Modelling 220(24) (2009), 3612–3620.

10.

K.K

, Liu

, Tang

J.J.

and Li

Y.W.

, Regional ecological security assessment along Anhui Section of Jiang-Ji-Huai Project based on pressure-state-response model, Journal of Anhui Jianzhu University 30(5) (2022), 1–9.

11.

Zhou

L.L.

, Wang

G.S.

, Yu

W.X.

and Tong

, Regional ecological security assessment based on pressure-state-response model: A case study of Enshi Prefecture, Hubei Province, Journal of Anhui Agricultural Sciences 44(18) (2016), 65–67, 75.

12.

Guo

and Wang

, Remote sensing monitoring and ecological risk assessment of landscape patterning in the agro-pastoral ecotone of northeast China, Complexity 2021 (2021), 8851543.

13.

Najafzadeh

and Basirian

, Evaluation of river water quality index using remote sensing and artificial intelligence models, Remote Sensing 15(9) (2023), 2359.

14.

Qiu

, Yang

, Zuo

, Wu

, Jiang

, Zhang

and Zhang

, Evaluation on the relevance of regional urbanization and ecological security in the nine provinces along the Yellow River, China, Ecological Indicators 132 (2021), 108346.

15.

Zhao

, Zhou

, Li

, Cheng

, Zhou

, Ma

, Li

and Huang

, Mapping urban dynamics – in Southeast Asia using consistent nighttime light data from DMSP and VIIRS, Remote Sensing of Environment 248 (2020), 111980.

16.

M.D.

, Li

, Peng

, Niu

and Cao

, Ecosystem health assessment based on RS and GIS, Ecology and Environmental Sciences 19(8) (2010), 1809–1814.

17.

Cao

M.L.

and Li

Y.D.

, Ecological security assessment of Tangshan City based on emergy analysis, Chinese Journal of Applied Ecology (9) (2009), 2214–2218.

18.

, Zhou

and Yi

, An integrated approach to constructing ecological security patterns and identifying ecological restoration and protection areas: A case study of Jingmen, China, Ecological Indicators 137 (2022), 108723.

19.

Quan

, Liang

, Yan

and Lei

, Influences of joint action of natural and social factors on atmosphericprocess of hydrological cycle in Inner Mongolia, China, Urban climate 41 (2022), 101043.

20.

Zheng

, Shen

and Zhong

, Construction of ecological security pattern in Pudacuo National Park, Acta Ecologica Sinica 41(3) (2021), 874–885.

21.

Ghosh

, Chatterjee

N.D.

and Dinda

, Urban ecological security assessment and forecasting using integrated DEMATEL-ANP and CA-Markov models: A case study on Kolkata Metropolitan Area, India, Sustainable Cities and Society 68 (2021), 102773.

22.

Yin

, Wei

C.F.

, Yang

X.Y.

and Luo

Y.J.

, The ecological compensation of land consolidation and its evaluation in hilly area of southwest China, Energy Procedia 5(22) (2011), 1192–1199.

23.

Liu

, Zhan

, Zhai

, Zhao

, Kumi

M.A.

, Wang

, Bai

and Wang

, Linking ecosystem service supply and demand to evaluate the ecological security in the Pearl River delta based on the pressure-state-response model, International Journal of Environmental Research and Public Health 20(5) (2023), 4062.

24.

, Wang

, Guo

, Yang

, Zhou

and Mougharbel

, Urban ecological security evaluation and spatial correlation research—–based on data analysis of 16 cities in Hubei Province of China, Journal of Cleaner Production 311 (2021), 127613.

25.

, Zhang

and Jia

, Humanization of nature: Testing the influences of urban park characteristics and psychological factors on collegers’ perceived restoration, Urban Forestry & Urban Greening 79 (2023), 127806.

26.

Zhang

X.Y.

, Wang

and Fan

Y.R.

, Feature selection using fuzzy-neighborhood relative decision entropy with class-level priority fusion, Journal of Intelligent & Fuzzy Systems 44(6) (2023), 9527–9544.

27.

Gao

Y.H.

, Wang

L.L.

and Zhang

H.Q.

, Intelligent urban ecological suitability system based on pattern recognition, Journal of Intelligent & Fuzzy Systems 39(4) (2020), 5009–5016.

28.

Wang

, Shi

, Zhang

, Tong

, Zhang

and Liu

, Research on landscape pattern construction and ecological restoration of Jiuquan City based on ecological security evaluation, Sustainability 13(10) (2021), 5732.

29.

Z.C.

, Liu

L.D.

, Sun

Y.F.

and Cui

J.R.

, Regional ecological security assessment based on PSR method, Acta Ecologica Sinica 30(23) (2010), 6495–6503.

Construction of ecological security evaluation model of healing landscape based on deep learning

Abstract

Keywords

1 Introduction

2 Research methods

2.1 Satellite data preprocessing

2.2 Classification of deep learning

2.3 Ecological security assessment

2.3.1 Selection of evaluation model

2.3.2 Principles of evaluation index system construction

2.3.3 Construction of evaluation index system

3 Case study

3.1 Overview of the study area

3.2 Data source and preprocessing

3.3 PSR model construction

3.3.1 Construction of landscape ecological security evaluation index

Table 2 Weight of ecological security assessment index system Target layer Criterion layer Index layer Weight w i A B1(0.2279) C1(0.0915) 0.4013 C2(0.1364) 0.5987 B2(0.4745) C3(0.2711) 0.5713 C4(0.0816) 0.2567 C5(0.1218) 0.1721 B3(0.2976) C6(0.1194) 0.4013 C7(0.1782) 0.5987

4.1 Precision of deep learning model

Table 4 Structure and parameter setting table of the model Network layer Settings Convolution Layer Conv2d(1,3,(3,51)) Attention Layer Conv2d(2, 3, (3, 51), padding = 0, bias = Flase); Sigmoid() Pooling Layer MaxPool2d((2, 1), 1) Fully Connected Layer Linear(51, 4) Classify Function Softmax

Footnotes

Funding

Conflicts of interest

References

Table 2
Weight of ecological security assessment index system

Target layer Criterion layer Index layer Weight w_i

A B1(0.2279) C1(0.0915) 0.4013

C2(0.1364) 0.5987

B2(0.4745) C3(0.2711) 0.5713

C4(0.0816) 0.2567

C5(0.1218) 0.1721

B3(0.2976) C6(0.1194) 0.4013

C7(0.1782) 0.5987

Table 4
Structure and parameter setting table of the model

Network layer Settings

Convolution Layer Conv2d(1,3,(3,51))

Attention Layer Conv2d(2, 3, (3, 51), padding = 0, bias = Flase); Sigmoid()

Pooling Layer MaxPool2d((2, 1), 1)

Fully Connected Layer Linear(51, 4)

Classify Function Softmax